Domain of the adjoint of an elliptic operator

184 Views Asked by At

I'm just trying to understand the concept of the adjoint of unbounded operators.

Let's look at the operator $Au:=-\epsilon\Delta u +(b\cdot\nabla)u+cu$
as an unbounded Operator on $L^{2}(\Omega)$ with $\epsilon>0$ and $b,c\in L^{\infty}(\Omega)$.

If we set $D(A):=H^{1}_{0}(\Omega)\cap H^{2}(\Omega)$ and the boundary of $\Omega$ is sufficently smooth, $A$ is a densely defined, closed operator.

Now, obviously the formal adjoint $F(A)$ of $A$ is $$F(A)u:=-\epsilon\Delta u -(b\cdot\nabla)u+cu$$ and we can give it the same Domain as $A$, i.e. $H^{1}_{0}(\Omega)\cap H^{2}(\Omega)$ to make it a densely defined closed Operator on $L^{2}(\Omega)$ with $$(Ax,y)_{L^{2}(\Omega)}=(x,F(A)y)_{L^{2}(\Omega)}$$ for all $x,y\in D(A)=D(F(A))$.

Now my question is: Is $A^{*}=F(A)$ with these definitions ? If so, why ? Why arent there any $y\in L^{2}(\Omega)\setminus D(A)$ so that $$(Ax,y)_{L^{2}(\Omega)}\leq K_{y} \|x\|_{L^{2}(\Omega)}$$ for all $x\in D(A)$ ?

Edit: I think I found an answer, though im not 100% convinced by it, because it seems to circumvent the condition $$y\in D(A^{*}) \Leftrightarrow (Ax,y)_{L^{2}(\Omega)}\leq K_{y} \|x\|_{L^{2}(\Omega)}\;\;\text{for all}\;\;x\in D(A)$$

Assertion: $D(A^{*})=D(A)$.
Proof: With the above mentioned condition and some integration by parts, it is easy to see, that $D(A)\subset D(A^{*})$. Let's show the other inclusion. For this we recall from standard theory of elliptic equations, that there exists a $\mu\in\mathbb{R}$, so that for all $\lambda\geq\mu$ the map $$A+\lambda I:D(A)\rightarrow L^{2}(\Omega)$$ is an isomorphism. Since $D(A+\lambda I)=D(A)$, we can assume from now on, that $A$ is a bijection. This is really the whole idea. On $D(A)$ we have $F(A)=A^{*}$ so $A^{*}$ is surjective. It is easy to see that $$\text{ker}(A^{*})=\text{im}(A)^{\bot}=L^{2}(\Omega)^{\bot}=\{0\}$$ so $A^{*}$ is also injective. If there was a $y\in L^{2}(\Omega)\setminus D(A)$ with $y\in D(A^{*})$, we would have an $x\in L^{2}(\Omega)$ and a $\hat{y}\in D(A)$ with $$A^{*}y=x=A^{*}\hat{y}$$ rendering $A^{*}$ not injective and proving the assertion by contradiction. $\square$

I am sorry about having posed this question and then shortly after coming up with an answer myself. I really did think about the problem for quite a few hours. I guess it helped me to formally explain my problem to others. If anyone finds a problem with my proof, please tell me ! I guess the updated question is: Is this proof correct ?